Adverse Weather Conditions Augmentation of LiDAR Scenes with Latent Diffusion Models

TL;DR

This paper introduces a latent diffusion model framework that generates realistic adverse weather LiDAR scenes from clear conditions, enhancing data diversity for autonomous driving systems.

Contribution

It proposes a novel latent diffusion process combined with autoencoder and postprocessing to generate realistic adverse weather LiDAR scenes from clear scenes.

Findings

Generated scenes improve robustness of autonomous driving models

Realism of adverse weather scenes is significantly enhanced

Method outperforms existing data augmentation techniques

Abstract

LiDAR scenes constitute a fundamental source for several autonomous driving applications. Despite the existence of several datasets, scenes from adverse weather conditions are rarely available. This limits the robustness of downstream machine learning models, and restrains the reliability of autonomous driving systems in particular locations and seasons. Collecting feature-diverse scenes under adverse weather conditions is challenging due to seasonal limitations. Generative models are therefore essentials, especially for generating adverse weather conditions for specific driving scenarios. In our work, we propose a latent diffusion process constituted by autoencoder and latent diffusion models. Moreover, we leverage the clear condition LiDAR scenes with a postprocessing step to improve the realism of the generated adverse weather condition scenes.